/dlg/ - Deep Learning General

deep learning what?

>>59983788
Every AI/ML thread on /g/ dies a painful death after 50 or so "hey I want to learn how do I start" posts by people that will never, ever understand the math behind ML

>>59984451
It's just a little linear algebra, statistics, and vector calc. :^)

>>59984487
It is, but they'll never get it anyway

4chan could use neural networks to auto-ban dumb frugposters

>>59984451
How accessible is it without the math? I want to make an image filter that gives anime girls Dorito hair.

File: Untitled.png (50KB, 958x688px) Image search: [Google]

50KB, 958x688px

DL + Laptop = Housefire

>>59984451
>>59984530

https://www.youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A/videos

>>59984530
Actually this would be more Computer Vision, a little less AI

https://www.youtube.com/watch?v=HgOV4yq8ZNw (not exactly what you're looking for, but same general idea)

https://www.youtube.com/watch?v=2S4nn7S8Hk4&list=PLAwxTw4SYaPnbDacyrK_kB_RUkuxQBlCm

>>59984684
He panders to normies too much. I doubt 90% of his viewership even understand what he's doing.

What the hell do you do when the model is training

any way to use ML in Unity? I'm not a great coder but i did a few courses on ML and my biggest wall is how to actually code the things rather than understanding how it works.

>>59986395
fap

>>59986919
Can GANs generate unlimited fap materials for me?

this is a sad thread

>>59986395
masturbate

does scikit-learn have neural networks built-in or do I have to use google's tensorflow? how do I into tensorflow if so?

>>59987219
use tensorflow or theano with keras, keras has the comfiest api currently

>>59987219
sklearn has no GPU support
Tensorflow is one of the few DL frameworks that support Windows, with Python 3.5 + Jupyter Notebook it's pretty easy to get started provided you have basic math background as detailed in>>59984487

>>59987339
yeah I'm well-versed in math so that shouldn't be an issue I just don't know python very well

>>59987383
sklearn has a multilayer perceptron if you wanna try doing that, its more straightforward than tensorflow, but if you wanna learn it right go with tensorflow

>>59987383
Do you know any other language? If so getting started is really quick. I came from C++/C#/Matlab and Python is one of the most intuitive language I've ever encountered

Look it up on Codecademy

>>59987419
I'll give it a try, I'm pretty familiar with C

>>59987473
then your completely fine, python is extremely straightforward to start, and has lots of little tricks once you start getting deeper into it

>>59987219
i've done some stats work for college with scikit and it was alright but theano is in another league. I don't think scikit can hold functions and give you its derivatives like theano does.

>>59983788
I see a vagina and a horse dildo. Anyone wanna give me a psyche evaluation?

what the fuck is going on in here

>>59988999
checked

and I'll have to get my Rorschach NN to think about that for you

File: 1491995336177.png (234KB, 882x758px) Image search: [Google]

234KB, 882x758px

>>59984558

>>59984684
POO

What are some good books I can read to start a decent career in data science + ML?

I need to stop being a brainlet :^(

>>59983788
deep "learning"/ANNs is/are cool, hip and all. problem is that it's nowadays used as go-to tool for solving problems, and not as something supplementary. people just "forgot" how use "classical" tools because of higher barrier of entry

>>59984487
CFD is just linear algebra, vector calc and little real analysis :^)

>>59984379
deep learn THIS *unzips*

>>59984487
this

>>59984487
I can do all of those things.

Explain to me what "deep learning" means.

>>59988999
Redditor.

>page 10
how can /g/ even compete?

>>59983788
It's not deep and not learning
>>59996058
Same shit as Dark matter
It's not dark and not matter, but we have to come up with something to explain some stuff we don't know/why our calculations don't work so here are some cool words!

>>59997827
autistic

File: 5deep6u.gif (890KB, 540x304px) Image search: [Google]

890KB, 540x304px

>>59999011
>Sayeth the one word responder

>>59986888
I never understood this meme
If you can't code it,you DON'T understand it
Don't kid yourself

>>59984487
someone post the meme with professor teaching algorithms and the CS student saying "What the hell I just wanted to learn to make video games"

File: 7102012232216iwsmt.jpg (82KB, 574x606px) Image search: [Google]

82KB, 574x606px

>>60001561
nvm found it

File: captcha.png (232KB, 403x524px) Image search: [Google]

232KB, 403x524px

I have an MNIST classifying neural network that I've trained for ~300,000 epochs. If I make another NN to take the probabilities output by the first one and train it to classify digits on that, will it learn to recognize the mistakes of my first NN and improve the accuracy (which is currently at 96%)

>>59984558

File: e53.png (1MB, 1552x1762px) Image search: [Google]

1MB, 1552x1762px

>>59984558

Jesus...

>>60001646
Yes, soft targets can be used for distillation and can be used as a regularizer as well

Most ml examples I've seen have focused on image recognition. But what can be done in terms of textual data? Let's say I want to match strings of text with each other based on the contents, how good would ml be at that?

>>60003829
very good
read about natural language processing

https://en.wikipedia.org/wiki/Outline_of_natural_language_processing#Natural_language_processing_toolkits

Redpill me

I want to create an AI and make them talk to each other

I don't even know where to begin I just have my riced out i3 laptop

Whats the difference between machine learning and deep learning?

>>60004003
>redpill me
You have seen the matrix right? it's literately kys softly with some sort of poison.

two processes, memory or file to 'talk' in. r/w handling so they don't talk over each other. Done!

>>60004130
deep learning is a subset of machine learning

>>60004148
But what is it exactly? What makes one algorithm just machine learning, and another one not just machine learning, but deep learning?

>>60004165
deep learning is still machine learning, so it can't be solely deep learning.
https://en.wikipedia.org/wiki/Deep_learning

>IT IS NOT THAT HARD

>>60004217
So it's basically a neural net with lots of layers?

>>60004295
And more blackboxed.

>>60003980
Do you know which one score the best in contextual understanding compared to a human baseline?